Making AI practical, human …. and accessible

 

The task we set for our participants during the session was, in a way, a mission impossible. We asked a diverse group of HR professionals, representing different fields and companies, to come up with a concrete mini-project that could be implemented in the short term as a shared effort of three.

 

Not a small ask to do in less than 90 minutes, considering their different contexts, priorities and realities.

 

What struck me was how hands-on the participants were. Many naturally gravitated towards processes and practicalities, or, how to get things done, how to make things work, rather than exploring from a helicopter view the role AI could play in their specific settings  in generic terms. That itself is an important insight and reveals the specialized expertise of HR professionals.

 

Before we can fully explore what AI might mean for HR and people in general, we need to make it tangible. We need to demystify AI, break it down into bite-sized pieces and connect it to the everyday challenges people face in their work. The focus should not be on AI as something that takes over, but on AI as something that helps people do things better, easier, or differently. Making a  good plan for the transition is key.

 

There is hope in this approach. For me, this is the way forward: start small, stay practical, and keep people at the centre. Rather than asking people to adapt to AI, let’s explore how AI can adapt to the needs of people. This is where HR can have a big role in my view.

 

Katja Einola

 

Download

Idea-generation

Post-read_Co-Creation-Labs_Katja-Einola+Violetta-Khoreva_idea-generation.pdf

 

Experiment

Post-read_Co-Creation-Labs_Katja-Einola+Violetta-Khoreva_experiment.pdf

 

Links

In this podcast, we share insights from our research on human-AI collaboration and discuss what HR leaders need to know about introducing AI successfully.

BoundarySpanningBookj-chapter2.pdf

 

Summary

How might we design work where AI helps us see talent beyond the pattern without diminishing human agency and judgement?

What the groups explored

Nearly all groups chose recruitment as the arena for their experiment. The interview was the most common focus: AI that helps prepare questions, suggests follow-up questions in real time, transcribes and summarises, or conducts a first screening interview. Others chose screening and matching of applicants, AI-supported search and outreach to candidates, or AI as a second opinion on the whole recruitment. Several groups specified where the person makes the call: AI suggests or recommends, and a recruiter or manager decides. Bias came up in several groups, both as something AI could help reduce and as a risk in the AI itself. Nearly all groups named GDPR, other legislation or compliance as a possible blocker, and several wanted the unions involved or informed. Several groups were also curious about how candidates would experience AI in the process.

First steps to test

• AI as support during the interview, with the recruiter in charge. The idea recurred in several forms. The interview is recorded, AI transcribes it in real time and suggests questions and follow-up questions based on the candidate’s answers, and the recruiter decides which questions to ask. AI can also help prepare and summarise the interview. Conditions named were the candidate’s consent, negotiation with the unions and training for recruiting managers and HR, with a pilot before any wider implementation. One suggestion was a cross-functional team from the start, including recruitment, HR, operations, legal and a sponsor in the management team.
• Interview questions built from the requirement profile. One group suggested letting AI draft interview questions from the requirement profile, to focus on personality and “soft values” and to avoid bias, using a recorded needs analysis as input.
• Automatic screening interviews, compared with traditional ones. One suggestion was that candidates do an automatic interview after applying, at a time that suits them, with questions set by the recruiter. The results are compared with a traditional interview, and a person decides who goes forward. What to follow: how candidates behave, who does not complete the interview, and differences in show-ups and no-shows. Clear instructions were named as a way to handle fear of AI. A related idea was an AI agent that conducts the interview and matches the answers against the role, with the question of how quickly and correctly it adapts to the specific organisation.
• Screening and outreach based on the organisation’s own requirements. One suggestion was to collect the organisation’s values, policies and the traits needed for a specific recruitment, let AI run this against the applications, have legal analyse the GDPR risks and inform union representatives. The group wanted to find out whether AI can be unbiased, whether their own judgement would be affected and whether candidates would notice. Another group suggested an AI workflow for search: match the requirement profile with potential candidates, review the longlist, draft a pitch, make initial contact and report back, while asking how far AI can go in contact with candidates before they no longer accept it.
• A second opinion on the whole recruitment. After the first interview, AI makes an overall assessment of every step: the job ad, screening questions, CVs, interview notes, requirement profile and interview guide. AI recommends and delivers material; the person decides and gives the candidate feedback. The group pointed out that interview transcripts may contain sensitive personal data, and that the review can also show where the wording of ads, profiles and questions should change.
• A structured review of AI output. Pick one prioritised process or project and set aside time to review what AI produced: did we ask the right questions, and should we move to a more focused pilot? Make sure key stakeholders understand the output and the way forward, keep a log of changes, and run an after action review to see where AI improves the quality of the output and where it does not.